MACHINE LEARNING APPLICATION FOR GAIT ABNORMALITY DETECTION USING INERTIAL MEASUREMENT UNITS

• Main Author: Angsetya, Raymond
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/67446
• Institution: Institut Teknologi Bandung

Disturbance on human gait can be used as indicator to detect a disease in human. Machine learning has been used to help in classifying diseases in medical science. The purpose of the application is to help doctor by detecting the diseases automatically. Previous researches tend to classify gait into healthy and one or few types of pathological gait. These automatic classification failed in task where detection of pathological gait in general is needed. Each of the model is limited to classify one or a few type of disease depending on the training dataset. Thus, the purpose of this research is to develop a method which able to detect gait abnormality in general. Data used in this research are taken from Inertial Measurement Unit. Linear acceleration and angular velocity data collected from the sensors will be standarized and alligned with allignment method called piecewise linear length normalization. Each of the gait signal is compared to the upper and lower control limit of gait signal for the purpose of calculating anomaly percentage of each gait cycle. The anomaly percentage of each signal is then fed to machine learning for training by random forest classifier. Best test result can be seen with sensors position are on the hip, knee, and feet, with accuracy value of 96.38% and recall value of 95.05%. The result of automatic detection is then completed with Shapley Additive Explanation to extract more information from the detection result for the doctor to do analysis.